feat: implement streaming methods for chat models

Author: radu-mocanu

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "uipath-langchain"
-version = "0.0.134"
+version = "0.0.135"
 description = "UiPath Langchain"
 readme = { file = "README.md", content-type = "text/markdown" }
 requires-python = ">=3.10"

src/uipath_langchain/_cli/_runtime/_conversation.py

@@ -173,6 +173,14 @@ def map_message(
                             content_part_sequence=idx,
                         ),
                     )
+        elif isinstance(message.content, str) and message.content:
+            msg_event.content_part = UiPathConversationContentPartEvent(
+                content_part_id=f"content-{message.id}",
+                chunk=UiPathConversationContentPartChunkEvent(
+                    data=message.content,
+                    content_part_sequence=0,
+                ),
+            )
 
         stop_reason = message.response_metadata.get("stop_reason")
         if not message.content and stop_reason in ("tool_use", "end_turn"):

src/uipath_langchain/chat/models.py

@@ -1,15 +1,15 @@
 import json
 import logging
-from typing import Any, Dict, List, Literal, Optional, Union
+from typing import Any, AsyncIterator, Dict, Iterator, List, Literal, Optional, Union
 
 from langchain_core.callbacks import (
     AsyncCallbackManagerForLLMRun,
     CallbackManagerForLLMRun,
 )
 from langchain_core.language_models import LanguageModelInput
-from langchain_core.messages import AIMessage, BaseMessage
+from langchain_core.messages import AIMessage, AIMessageChunk, BaseMessage
 from langchain_core.messages.ai import UsageMetadata
-from langchain_core.outputs import ChatGeneration, ChatResult
+from langchain_core.outputs import ChatGeneration, ChatGenerationChunk, ChatResult
 from langchain_core.runnables import Runnable
 from langchain_openai.chat_models import AzureChatOpenAI
 from pydantic import BaseModel
@@ -49,6 +49,54 @@ async def _agenerate(
         response = await self._acall(self.url, payload, self.auth_headers)
         return self._create_chat_result(response)
 
+    def _stream(
+        self,
+        messages: List[BaseMessage],
+        stop: Optional[List[str]] = None,
+        run_manager: Optional[CallbackManagerForLLMRun] = None,
+        **kwargs: Any,
+    ) -> Iterator[ChatGenerationChunk]:
+        if "tools" in kwargs and not kwargs["tools"]:
+            del kwargs["tools"]
+        payload = self._get_request_payload(messages, stop=stop, **kwargs)
+        response = self._call(self.url, payload, self.auth_headers)
+
+        # For non-streaming response, yield single chunk
+        chat_result = self._create_chat_result(response)
+        chunk = ChatGenerationChunk(
+            message=AIMessageChunk(
+                content=chat_result.generations[0].message.content,
+                additional_kwargs=chat_result.generations[0].message.additional_kwargs,
+                response_metadata=chat_result.generations[0].message.response_metadata,
+                usage_metadata=chat_result.generations[0].message.usage_metadata,  # type: ignore
+            )
+        )
+        yield chunk
+
+    async def _astream(
+        self,
+        messages: List[BaseMessage],
+        stop: Optional[List[str]] = None,
+        run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
+        **kwargs: Any,
+    ) -> AsyncIterator[ChatGenerationChunk]:
+        if "tools" in kwargs and not kwargs["tools"]:
+            del kwargs["tools"]
+        payload = self._get_request_payload(messages, stop=stop, **kwargs)
+        response = await self._acall(self.url, payload, self.auth_headers)
+
+        # For non-streaming response, yield single chunk
+        chat_result = self._create_chat_result(response)
+        chunk = ChatGenerationChunk(
+            message=AIMessageChunk(
+                content=chat_result.generations[0].message.content,
+                additional_kwargs=chat_result.generations[0].message.additional_kwargs,
+                response_metadata=chat_result.generations[0].message.response_metadata,
+                usage_metadata=chat_result.generations[0].message.usage_metadata,  # type: ignore
+            )
+        )
+        yield chunk
+
     def with_structured_output(
         self,
         schema: Optional[Any] = None,
@@ -217,6 +265,92 @@ async def _agenerate(
         response = await self._acall(self.url, payload, self.auth_headers)
         return self._create_chat_result(response)
 
+    def _stream(
+        self,
+        messages: List[BaseMessage],
+        stop: Optional[List[str]] = None,
+        run_manager: Optional[CallbackManagerForLLMRun] = None,
+        **kwargs: Any,
+    ) -> Iterator[ChatGenerationChunk]:
+        """Stream the LLM on a given prompt.
+
+        Args:
+            messages: the prompt composed of a list of messages.
+            stop: a list of strings on which the model should stop generating.
+            run_manager: A run manager with callbacks for the LLM.
+            **kwargs: Additional keyword arguments.
+
+        Returns:
+            An iterator of ChatGenerationChunk objects.
+        """
+        if kwargs.get("tools"):
+            kwargs["tools"] = [tool["function"] for tool in kwargs["tools"]]
+        if "tool_choice" in kwargs and kwargs["tool_choice"]["type"] == "function":
+            kwargs["tool_choice"] = {
+                "type": "tool",
+                "name": kwargs["tool_choice"]["function"]["name"],
+            }
+        payload = self._get_request_payload(messages, stop=stop, **kwargs)
+        response = self._call(self.url, payload, self.auth_headers)
+
+        # For non-streaming response, yield single chunk
+        chat_result = self._create_chat_result(response)
+        chunk = ChatGenerationChunk(
+            message=AIMessageChunk(
+                content=chat_result.generations[0].message.content,
+                additional_kwargs=chat_result.generations[0].message.additional_kwargs,
+                response_metadata=chat_result.generations[0].message.response_metadata,
+                usage_metadata=chat_result.generations[0].message.usage_metadata,  # type: ignore
+                tool_calls=getattr(
+                    chat_result.generations[0].message, "tool_calls", None
+                ),
+            )
+        )
+        yield chunk
+
+    async def _astream(
+        self,
+        messages: List[BaseMessage],
+        stop: Optional[List[str]] = None,
+        run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
+        **kwargs: Any,
+    ) -> AsyncIterator[ChatGenerationChunk]:
+        """Async stream the LLM on a given prompt.
+
+        Args:
+            messages: the prompt composed of a list of messages.
+            stop: a list of strings on which the model should stop generating.
+            run_manager: A run manager with callbacks for the LLM.
+            **kwargs: Additional keyword arguments.
+
+        Returns:
+            An async iterator of ChatGenerationChunk objects.
+        """
+        if kwargs.get("tools"):
+            kwargs["tools"] = [tool["function"] for tool in kwargs["tools"]]
+        if "tool_choice" in kwargs and kwargs["tool_choice"]["type"] == "function":
+            kwargs["tool_choice"] = {
+                "type": "tool",
+                "name": kwargs["tool_choice"]["function"]["name"],
+            }
+        payload = self._get_request_payload(messages, stop=stop, **kwargs)
+        response = await self._acall(self.url, payload, self.auth_headers)
+
+        # For non-streaming response, yield single chunk
+        chat_result = self._create_chat_result(response)
+        chunk = ChatGenerationChunk(
+            message=AIMessageChunk(
+                content=chat_result.generations[0].message.content,
+                additional_kwargs=chat_result.generations[0].message.additional_kwargs,
+                response_metadata=chat_result.generations[0].message.response_metadata,
+                usage_metadata=chat_result.generations[0].message.usage_metadata,  # type: ignore
+                tool_calls=getattr(
+                    chat_result.generations[0].message, "tool_calls", None
+                ),
+            )
+        )
+        yield chunk
+
     def with_structured_output(
         self,
         schema: Optional[Any] = None,